[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d4e30634-b64e-47c7-9089-a37d20e29d2f@kernel.org>
Date: Thu, 15 May 2025 10:17:05 +0100
From: Quentin Monnet <qmo@...nel.org>
To: Jiayuan Chen <jiayuan.chen@...ux.dev>, bpf@...r.kernel.org
Cc: Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>, Andrii Nakryiko <andrii@...nel.org>,
Martin KaFai Lau <martin.lau@...ux.dev>, Eduard Zingerman
<eddyz87@...il.com>, Song Liu <song@...nel.org>,
Yonghong Song <yonghong.song@...ux.dev>,
John Fastabend <john.fastabend@...il.com>, KP Singh <kpsingh@...nel.org>,
Stanislav Fomichev <sdf@...ichev.me>, Hao Luo <haoluo@...gle.com>,
Jiri Olsa <jolsa@...nel.org>, Daniel Xu <dxu@...uu.xyz>,
Mykyta Yatsenko <yatsenko@...a.com>, Tao Chen <chen.dylane@...il.com>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH bpf-next v2] bpftool: Add support for custom BTF path in
prog load/loadall
2025-05-15 14:50 UTC+0800 ~ Jiayuan Chen <jiayuan.chen@...ux.dev>
> This patch exposes the btf_custom_path feature to bpftool, allowing users
> to specify a custom BTF file when loading BPF programs using prog load or
> prog loadall commands.
>
> The argument 'btf_custom_path' in libbpf is used for those kernes that
Typo: "kernes"
> don't have CONFIG_DEBUG_INFO_BTF enabled but still want to perform CO-RE
> relocations.
>
> Signed-off-by: Jiayuan Chen <jiayuan.chen@...ux.dev>
> ---
> tools/bpf/bpftool/Documentation/bpftool-prog.rst | 7 ++++++-
> tools/bpf/bpftool/bash-completion/bpftool | 2 +-
> tools/bpf/bpftool/prog.c | 12 +++++++++++-
> 3 files changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/tools/bpf/bpftool/Documentation/bpftool-prog.rst b/tools/bpf/bpftool/Documentation/bpftool-prog.rst
> index d6304e01afe0..e60a829ab8d0 100644
> --- a/tools/bpf/bpftool/Documentation/bpftool-prog.rst
> +++ b/tools/bpf/bpftool/Documentation/bpftool-prog.rst
> @@ -127,7 +127,7 @@ bpftool prog pin *PROG* *FILE*
> Note: *FILE* must be located in *bpffs* mount. It must not contain a dot
> character ('.'), which is reserved for future extensions of *bpffs*.
>
> -bpftool prog { load | loadall } *OBJ* *PATH* [type *TYPE*] [map { idx *IDX* | name *NAME* } *MAP*] [{ offload_dev | xdpmeta_dev } *NAME*] [pinmaps *MAP_DIR*] [autoattach]
> +bpftool prog { load | loadall } *OBJ* *PATH* [type *TYPE*] [map { idx *IDX* | name *NAME* } *MAP*] [{ offload_dev | xdpmeta_dev } *NAME*] [pinmaps *MAP_DIR*] [autoattach] [kernel_btf *BTF_DIR*]
> Load bpf program(s) from binary *OBJ* and pin as *PATH*. **bpftool prog
> load** pins only the first program from the *OBJ* as *PATH*. **bpftool prog
> loadall** pins all programs from the *OBJ* under *PATH* directory. **type**
> @@ -153,6 +153,11 @@ bpftool prog { load | loadall } *OBJ* *PATH* [type *TYPE*] [map { idx *IDX* | na
> program does not support autoattach, bpftool falls back to regular pinning
> for that program instead.
>
> + The **kernel_btf** option allows specifying an external BTF file to replace
> + the system's own vmlinux BTF file for CO-RE relocations. NOTE that any
> + other feature (e.g., fentry/fexit programs, struct_ops, etc) will require
Nit: No need for both "e.g." and "etc", they're redundant.
> + actual kernel BTF like /sys/kernel/btf/vmlinux.
> +
Can we rephrase the second part of the paragraph a little bit please?
“Any other feature” could be clearer, how about:
Note that any other feature relying on BTF (such as fentry/fexit
programs, struct_ops) requires the BTF file for the actual
kernel running on the host, often exposed at
/sys/kernel/btf/vmlinux.
> Note: *PATH* must be located in *bpffs* mount. It must not contain a dot
> character ('.'), which is reserved for future extensions of *bpffs*.
>
> diff --git a/tools/bpf/bpftool/bash-completion/bpftool b/tools/bpf/bpftool/bash-completion/bpftool
> index 1ce409a6cbd9..609938c287b7 100644
> --- a/tools/bpf/bpftool/bash-completion/bpftool
> +++ b/tools/bpf/bpftool/bash-completion/bpftool
> @@ -511,7 +511,7 @@ _bpftool()
> ;;
> *)
> COMPREPLY=( $( compgen -W "map" -- "$cur" ) )
> - _bpftool_once_attr 'type pinmaps autoattach'
> + _bpftool_once_attr 'type pinmaps autoattach kernel_btf'
> _bpftool_one_of_list 'offload_dev xdpmeta_dev'
> return 0
> ;;
Correct, but right before this could you also add the following, please:
@@ -505,13 +505,13 @@ _bpftool()
_bpftool_get_map_names
return 0
;;
- pinned|pinmaps)
+ pinned|pinmaps|kernel_btf)
_filedir
return 0
;;
*)
This will make the completion offer file names after the user has typed
"kernel_btf".
> diff --git a/tools/bpf/bpftool/prog.c b/tools/bpf/bpftool/prog.c
> index f010295350be..3b6a361dd0f8 100644
> --- a/tools/bpf/bpftool/prog.c
> +++ b/tools/bpf/bpftool/prog.c
> @@ -1681,8 +1681,17 @@ static int load_with_options(int argc, char **argv, bool first_prog_only)
> } else if (is_prefix(*argv, "autoattach")) {
> auto_attach = true;
> NEXT_ARG();
> + } else if (is_prefix(*argv, "kernel_btf")) {
> + NEXT_ARG();
> +
> + if (!REQ_ARGS(1))
> + goto err_free_reuse_maps;
> +
> + open_opts.btf_custom_path = GET_ARG();
> } else {
> - p_err("expected no more arguments, 'type', 'map' or 'dev', got: '%s'?",
> + p_err("expected no more arguments, "
> + "'type', 'map', 'dev', 'offload_dev', 'xdpmeta_dev', 'pinmaps', "
> + "'autoattach', or 'kernel_btf', got: '%s'?",
Some of them were missing, thanks for this! Can you remove "dev" from
the list, please? It's been deprecated in favour of "offload_dev", to
avoid confusion with "xdpmeta_dev".
pw-bot: cr
> *argv);
> goto err_free_reuse_maps;
> }
> @@ -2474,6 +2483,7 @@ static int do_help(int argc, char **argv)
> " [map { idx IDX | name NAME } MAP]\\\n"
> " [pinmaps MAP_DIR]\n"
> " [autoattach]\n"
> + " [kernel_btf BTF_DIR]\n"
> " %1$s %2$s attach PROG ATTACH_TYPE [MAP]\n"
> " %1$s %2$s detach PROG ATTACH_TYPE [MAP]\n"
> " %1$s %2$s run PROG \\\n"
Thanks,
Quentin
Powered by blists - more mailing lists